Paradox of Supervision
- Entity ID:
ent-20260423-r31a000000012 - Type:
concept - Scope:
private - Status:
active
Description
The principle — documented by Anthropic's 132-engineer internal study — that AI assistance risks atrophying the skills required to supervise the AI. Central to the VILA paper's sixth 'evaluative lens' (long-term human capability preservation) which they identify as absent from Anthropic's stated design values.
Key claims
- Anthropic's 132-engineer internal study documents a 'paradox of supervision'
- AI-assisted developers score 17% lower on code comprehension tests
Relations
- Anthropic 132-Engineer Internal Study (Paradox of Supervision) --[supports]--> Paradox of Supervision
- 17% Code Comprehension Reduction (AI-Assisted Developers) --[supports]--> Paradox of Supervision
- Paradox of Supervision --[motivates]--> Future Direction: Long-Term Capability Preservation as First-Class Design Problem